Search Images Maps Play YouTube News Gmail Drive More »
Sign in
Screen reader users: click this link for accessible mode. Accessible mode has the same essential features but works better with your reader.

Patents

  1. Advanced Patent Search
Publication numberUS6055615 A
Publication typeGrant
Application numberUS 08/792,134
Publication dateApr 25, 2000
Filing dateJan 31, 1997
Priority dateFeb 2, 1996
Fee statusPaid
Also published asDE69725632D1, DE69725632T2, EP0788110A2, EP0788110A3, EP0788110B1, US6163832, US6507900
Publication number08792134, 792134, US 6055615 A, US 6055615A, US-A-6055615, US6055615 A, US6055615A
InventorsYoshinori Okajima
Original AssigneeFujitsu Limited
Export CitationBiBTeX, EndNote, RefMan
External Links: USPTO, USPTO Assignment, Espacenet
Pipeline memory access using DRAM with multiple independent banks
US 6055615 A
Abstract
A memory system using at least one DRAM chip and equipped with an interface for transferring input/output data in a packet format includes a plurality of banks within each of the at least one DRAM chip. The memory system further includes a control circuit for accessing a bank for data transfer of a given packet when the bank is different from a previous bank accessed for an immediately preceding packet, and for waiting for an operation to complete in the bank when the bank is the same as the previous bank.
Images(17)
Previous page
Next page
Claims(6)
What is claimed is:
1. A memory system using at least one DRAM chip, said memory system comprising:
a first bus for transferring an address signal and a control signal as a packet;
a second bus for transferring input/output data;
a plurality of banks within each of said at least one DRAM chip; and
a control circuit for accessing a bank for data transfer of a given packet when said bank is different from a previous bank accessed for an immediately preceding packet, and for waiting for an operation to complete in said bank when said bank is the same as said previous bank, wherein each of said at least one DRAM chip comprises a first interface circuit coupled to said first bus for receiving said packet, and a second interface circuit coupled to said second bus.
2. A memory system using at least one DRAM chip, said memory system comprising:
a first bus for transferring an address signal and a control signal as a packet;
a second bus for transferring input/output data;
a plurality of banks within each of said at least one DRAM chip; and
a control circuit for accessing a bank for data transfer of a given packet when said bank is different from any of previous banks accessed in preceding predetermined time period, and for waiting for an operation to complete in said bank when said bank is the same as one of said previous banks, wherein each of said at least one DRAM chip comprises a first interface circuit coupled to said first bus for receiving said packet, and a second interface circuit coupled to said second bus.
3. The memory system as claimed in claim 1, wherein access to said at least one DRAM chip is made by a command packet which is formed by multiplexing a device address, a row address, a column address, a bank address, and a command, wherein said device address indicates one of said at least one DRAM chip.
4. The memory system as claimed in claim 2, wherein access to said at least one DRAM chip is made by a command packet which is formed by multiplexing a device address, a row address, a column address, a bank address, and a command, wherein said device address indicates one of said at least one DRAM chip.
5. The memory system as claimed in claim 3, wherein said at least one DRAM chip receives a one-bit signal in addition to said command packet, said one-bit signal indicates a start of said command.
6. The memory system as claimed in claim 4, wherein said at least one DRAM chip receives a one-bit signal in addition to said command packet, said one-bit signal indicates a start of said command.
Description
BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention generally relates to pipeline-operation devices, and particularly relates to a memory device carrying out pipeline operations.

2. Description of the Prior Art

As a foundation of an information society, DRAM (dynamic random access memory) chips are widely used because of their capacity to allow integrated circuits to be manufactured in high density. In order to boost speed of data-read/write operations, DRAMs are typically provided with various functions contrived for the purpose of speed enhancement, an example of such functions being a page mode. With an aim of achieving high-speed data transfer, also, SDRAM (synchronous DRAM) is created as a variation of DRAMs to carry out data-read/write operations in synchronism with a clock-signal.

FIGS. 1A and 1B are time charts showing a data-read timing of a conventional DRAM operating in the page mode. FIGS. 1C through 1E are time charts showing a data-read timing of an SDRAM operating in a burst mode.

In the DRAM page mode as shown in FIG. 1A, a plurality of column addresses CA1, CA2, CA3, and CA4 are provided during a single cycle between an input of a given row address RA1 and an input of a next row address RA2. As shown in FIG. 1B, four pieces of data Q1, Q2, Q3, and Q4 in this case can be read out at 20-ns intervals, for example, as long as these pieces of data are stored in the same row address. When data is stored in consecutive addresses or in the same row address, use of the page mode is highly effective.

In the SDRAM burst mode as shown in FIG. 1D, a first column address CA1 is provided in synchronism with a clock signal CLK (FIG. 1C) during one cycle between an input of a given row address RA1 and an input of a next row address RA2, and column addresses following the first column address CA1 are internally generated in the memory chip. As shown in FIG. 1E, data pieces Q1, Q2, Q3, and Q4 are continuously read from the memory at high speed in synchronism with the clock signal CLK, as long as these data pieces are stored in the same row address. Similar to the requirements of the DRAM page mode, the SDRAM burst mode achieves a high-speed-data-read operation only when data addresses are continuous or have the same row address.

The DRAM page mode of FIG. 1A and the SDRAM burst mode of FIG. 1D have the same cycle period between successive row-address inputs as that of DRAMs operating in the normal mode. Because of this, the use of the SDRAM or the DRAM page mode can achieve only a data-read speed as fast as that of the normal mode when the data-read addresses are random rather than in the same row address.

DRAMs require a sequence of operations to be conducted within one cycle period between successive row accesses, such sequence including precharging bit-lines (column addresses) to get the bit-lines ready, selecting a word-line (row address) to transfer data to a sense amplifier, and selecting a bit-line to read the data. Because of such a long sequence, the cycle of the row accesses becomes as much as about 100 ns. In other words, since DRAMs require the precharging of bit-lines prior to each data-read operation, consecutive data-read operations are difficult to achieve when read-data addresses are random.

In order to overcome this difficulty, a memory may be divided into a plurality of banks. In principle, this is the same as providing a plurality of memory chips of the same type. With N banks provided, random data access should end up accessing a bank different from the one accessed immediately before at a probability of (N-1)/N. Each bank other than the bank accessed immediately before is already precharged and ready, so that consecutive data-read/write operations can be achieved by accessing different banks one after another.

In such a memory-bank method, however, each bank should be provided with a dedicated set of control circuits while only one of such a set is normally required for a whole memory device. This means that an increase in the bank number leads to an enlargement in the chip size.

Accordingly, there is a need for a memory device which allows address accesses to be consecutively made at high speed by providing a plurality of memory blocks capable of identical operations without enlarging the chip size.

SUMMARY OF THE INVENTION

Accordingly, it is a general object of the present invention to provide a memory device which can satisfy the need described above.

It is another and more specific object of the present invention to provide a memory device which allows address accesses to be consecutively made at high speed by providing a plurality of memory blocks capable of identical operations without enlarging the chip size.

In order to achieve the above objects according to the present invention, a memory system using at least one DRAM chip and equipped with an interface for transferring input/output data in a packet format includes a plurality of banks within each of the at least one DRAM chip. The memory system further includes a control circuit for accessing a bank for data transfer of a given packet when the bank is different from a previous bank accessed for an immediately preceding packet, and for waiting for an operation to complete in the bank when the bank is the same as the previous bank.

In the device described above, the control circuit for selecting a bank is shared by the plurality of banks so that there is no need to provide a complex controlling mechanism for each of the banks. In addition, while a given bank is carrying out a given operation, another bank which has already finished this given operation can carry out a next operation following the given operation. In this manner, the banks as a whole perform predetermined operations in a pipe-line manner, thereby achieving high-speed processing. Further, when the same bank is successively accessed, an access made at a later timing waits for the operation of the previous access to finish in the bank, thereby providing a mechanism for coping with a disturbance of the pipe-line operation.

Other objects and further features of the present invention will be apparent from the following detailed description when read in conjunction with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

FIGS. 1A and 1B are time charts showing a data-read timing of a conventional DRAM operating in a page mode;

FIGS. 1C through 1E are time charts showing a data-read timing of an SDRAM operating in a burst mode;

FIGSs. 2A and 2B are block diagrams showing configurations of a pipeline-operation device according to a principle of the present invention;

FIG. 3 is a block diagram of DRAM according to a first embodiment of the principle of the present invention;

FIGS. 4A through 4L are illustrative drawings showing schedules of blocks of the DRAM according to the present invention;

FIGS. 5A and 5B are table charts showing con tents of a command/address input;

FIGS. 6A through 6H are time charts showing relations between a Flag-0 input, the command/address input, and other signals;

FIGS. 7A through 7D are time charts showing a strobe signal, an internal clock Φ4, an internal clock Θ1, and an inverted internal clock /Θ1, respectively;

FIG. 8 is a circuit diagram showing a partial configuration of the global row scheduler of FIG. 3;

FIG. 9 is a circuit diagram showing a partial configuration of the global row scheduler of FIG. 3;

FIG. 10 is a circuit diagram of the column scheduler of FIG. 3;

FIG. 11 is a circuit diagram of the local row scheduler of FIG. 3;

FIG. 12 is a circuit diagram of a circuit for a parallel-to-serial conversion for chip output;

FIGS. 13A through 13E are time charts of signals used in the circuit of FIG. 12; and

FIGS. 14A through 14L are illustrative drawings showing schedules of blocks of an SDRAM according to the present invention when the burst length is 1 and the CAS latency is 1.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

In the following, a principle and embodiments of the present invention will be described with reference to the accompanying drawings.

FIGS. 2A and 2B are block diagrams showing configurations of a pipeline-operation device according to a principle of the present invention.

The pipeline-operation device of FIG. 2A according to the principle of the present invention includes a control device 1, a plurality of blocks (banks) 2, and a control line 3 connecting the control device 1 with each block 2. Each of the blocks 2 carries out M different operations Al through AM in a predetermined order. The control device 1 selects one of the plurality of blocks 2 via the control line 3. The selected block 2 carries out the M operations in the predetermined order. A time length required by each of the M operations defines one cycle, and the control device 1 selects one block in each cycle from the blocks 2 which are not in operation at the time of the selection. The M operations of the selected block 2 will finish after M cycles, so that the number of the selected blocks 2 simultaneously in operation are M.

By nature of the above operation arrangement, the operation AM, for example, should be being carried out at a given cycle by one of the plurality of blocks 2. Although the operation AM of one block should be preceded by the M-1 operations A1 through AM-1 by the same block, the plurality of the blocks 2 in its entirety can carry out the operation AM in each cycle. That is, the plurality of the blocks 2 in its entirety are capable of a pipeline-like operation.

FIG. 2B shows an alternate example of a pipeline-operation device according to the principle of the present invention. The pipeline-operation device of FIG. 2B includes a control device 1A, a plurality of blocks (banks) 2A, and M (three in the figure) control lines 3A connecting the control device 1A with each of the blocks 2A. Each of the plurality of blocks 2A carries out the different operations A1 through AM corresponding in number to the control lines 3A. The control device 1A selects one of the blocks 2A, and uses the control lines 3A to indicate one of the M operations which is to be carried out by the selected block 2A. In this manner, the selected block 2A carries out the indicated operation. A time length required for the indicated operation defines one cycle, and the control device 1A selects one of the M control lines 3A for the selected block 2A in each cycle, such that the selected block 2A carries out the M operations in the predetermined order. Also, the control device 1A selects one new block in each cycle from the blocks 2A which are not in operation at the time of the selection, and makes the newly selected block 2A carry out the M operations in the predetermined order in the same manner as for the already selected blocks 2A. The M operations of the selected block 2A will finish after M cycles, so that the M selected blocks 2A are simultaneously in operation.

Accordingly, in the same manner as in the configuration of FIG. 2A, the operation AM, for example, should be being carried out at a given cycle by one of the plurality of blocks 2A. Although the operation AM of one block should be preceded by the M-1 operations Al through AM-1 by the same block, the plurality of the blocks 2A in their entirety can carry out the operation AM in each cycle. That is, the plurality of the blocks 2A in their entirety are capable of a pipeline-like operation.

In the configuration of FIG. 2A, each block 2 should be equipped with a function to carry out the M operations in the predetermined order, but the number of the control lines between the control device 1 and the blocks 2 are fewer than that of FIG. 2B. In FIG. 2B, each block 2A, since it is required to execute only an indicated operation, has a simpler structure than that of the block 2 of FIG. 2A, but the lines connecting the blocks 2A and the control device 1A become complex. Here, the function included in the block 2 of FIG. 2A to execute the M operations in the predetermined order is basically the same as a function provided in the control device 1A of FIG. 2B to control each block 2A to carry out the M operations in the predetermined order. That is, the difference between the configuration of FIG. 2A and the configuration of FIG. 2B is whether to place the function of executing the M operations in the predetermined order in each block or to collectively put this function in the control device.

FIG. 3 is a block diagram of DRAM according to a first embodiment of the principle of the present invention. The embodiment of the DRAM of FIG. 3 corresponds to the configuration shown in FIG. 2A, but can be easily modified to correspond to the configuration of FIG. 2B. The DRAM of FIG. 3 according to the principle of the present invention includes an input circuit 10, a synchronization-control circuit 20, a data-input/output circuit 30, a global row scheduler 41, a refresh-address counter 42, a read/write scheduler 43, a column scheduler 44, a burst-address counter 45, a data-transfer/column decoder 46, and a plurality of blocks 50. In FIG. 3, the global row scheduler 41 corresponds to the control device 1 of FIG. 2A.

The input circuit 10 receives an input signal at a plurality of input nodes IN (only one shown in FIG. 3) of the DRAM. Then, the input circuit 10 extracts, from the input signal, address data regarding an address of a data-read/write operation, and command data regarding a command given to each of the blocks 50. The input circuit 10 includes an input-signal buffer 11 for temporal storage of the input signal, an address register 12, an address pre-decoder 13, a command register 14, a command decoder 15, a mode register 16, and a mode decoder 17.

The address register 12 receives the input signal supplied from the input-signal buffer 11, and extracts a read/write address from the input signal for storage therein. The address pre-decoder 13 decodes the address provided from the address register 12. The command register 14 receives the input signal supplied from the input-signal buffer 11, and extracts a command indicating an operation of each of the blocks 50 to store the command therein. The command decoder 15 decodes the command supplied from the command register 14. The mode register 16 stores mode setting data in order to set an operation latency, a refresh mode, etc., in the same manner as in a conventional DRAM. The mode decoder 17 decodes the mode supplied from the mode register 16.

The address, the mode, and the command sent from the input circuit 10 are supplied to each of the global row scheduler 41, the read/write scheduler 43, and the column scheduler 44. The global row scheduler 41, the read/write scheduler 43, and the column scheduler 44 are shared by the plurality of the blocks 50.

The synchronization-control circuit 20 generates an internal clock based on a strobe signal and a control signal including a flag signal to be described later, where the strobe signal is input to an input node Strobe of the DRAM, and the control signal is input at a control-signal input node CTRL of the DRAM. The synchronization-control circuit 20 includes a strobe buffer 21 for a temporal storage of the strobe signal, a control-signal buffer 22 for a temporal storage of the control signal, and an internal-clock generator 23 for generating internal clocks based on the strobe signal and the control signal. The generated internal clock is supplied to the global row scheduler 41, the input circuit 10, the read/write scheduler 43, the column scheduler 44, and the data-input/output circuit 30.

The data-input/output circuit 30 is responsible for the signal input and the signal output via the data-input/output node D/Q of the DRAM. The data-input/output circuit 30 includes an input/output buffer 31 for a temporal storage of input/output data for data input/output via the data-input/output node D/Q, and an input/output register 32 for storing the input/output data based on the internal clock and the read/write operations.

Each of the blocks 50 is a data storage unit which includes a memory-cell array for storing data, various drivers, a register for storing a row address, and a circuit for controlling executions of various operations in a predetermined order. Each of the blocks 50 includes a local row scheduler 51 for executing each operation of the pertinent block in the predetermined order, a row-address register 52 for storing a supplied row address, word-line drivers 53 and 54 for selecting a word-line indicated by the row address, a sense-amplifier driver 55 for driving a sense amplifier, memory arrays 56 and 57, and a sense amplifier 58. The sense amplifier 58 of each block 50 is connected to a bus 60 which is a shared two-way-communication path. The bus 60 conducts data transfer between the sense amplifier 58 and the data-transfer/column decoder 46.

Every single one of the banks provided in the prior-art SDRAM has the data-transfer/column decoder 46. In the present invention, however, the data-transfer/column decoder 46 and the bus 60 are shared by the plurality of the blocks 50. The difference between the block of the present invention and the bank of the prior-art SDRAM relies on this fact. Because of this fact, the present invention can reduce the chip area size compared to the prior art.

The refresh-address counter 42 generates row addresses one after another when each of the blocks 50 carries out a refresh operation. The generated row addresses are supplied to the global row scheduler 41. The burst-address counter 45 generates column addresses one after another in the burst mode when data is read from consecutive column addresses in the same row address. The generated column address is supplied to the column scheduler 44.

The global row scheduler 41 receives the address and the command from the input circuit 10 and the internal clock from the synchronization-control circuit 20, and makes a schedule of row-address accesses for each of the blocks 50. Namely, the global row scheduler 41 extracts a block address and a row address from the received address, selects one of the blocks 50 indicated by the block address, and instructs the selected one of the blocks 50 to chose a word-line indicated by the row address. Also, the global row scheduler 41 responsive to the received command instructs the selected one of the blocks 50 to execute operations such as a precharge operation, a refresh operation, etc. Such instructions are given in accordance with the command and the internal clock, as will be described later.

The column scheduler 44 receives the address and the command from the input circuit 10 and the internal clock from the synchronization-control circuit 20, and makes a schedule of column-address accesses for each of the blocks 50. Namely, the column scheduler 44 extracts the column address from the received address, and supplies the column address to the data-transfer/column decoder 46. The timing of the provision of the column address is determined based on the command and the internal clock. In this manner, the data-transfer/column decoder 46 can select a sense amplifier connected to a bit-line corresponding to the column address at the time of the data-read/write operation.

The read/write scheduler 13 receives the address and the command from the input circuit 10 and the internal clock from the synchronization-control circuit 20, and makes a schedule of data-read/write operations for each of the blocks 50. Namely, the read/write scheduler 43 extracts the block address from the received address, selects one of the blocks 50 indicated by the block address, and drives the sense-amplifier driver 55 of the selected one of the blocks 50. This timing is determined based on the command and the internal clock. Also, the read/write scheduler 43 notifies the data-transfer/column decoder 46 of the timing of the data-read/write operations.

FIGS. 4A through 4L are illustrative drawings showing schedules of blocks of the DRAM according to the present invention. FIG. 4A shows command inputs, and FIG. 4F is a continuation of FIG. 4A. FIGS. 4B through 4D show operation schedules of blocks 1 through 3, respectively, and FIGS. 4G through 4I are continuations of FIGS. 4B through 4D, respectively. FIGS. 4J and 4K show operation schedules of blocks 4 and 5, respectively. FIG. 4E shows data outputs, and FIG. 4L is a continuation of FIG. 4E.

As shown in FIGS. 4A through 4L, one of the blocks 50 (FIG. 3) selected by the global row scheduler 41 (FIG. 3) can execute one command during four cycles of the clock (strobe) signal. Each block operates such that a state thereof switches once every four cycles. FIGS. 4A through 4L show a case in which the burst length is 4. In this case, 4-bit data is read out at a time in parallel to be converted into serial data, and this operation is executed two times for each block access.

With reference to FIG. 3 and FIGS. 4A through 4L, each block is initially in a ready condition. When the global row scheduler 41 selects one of the blocks 50, the local row scheduler 51 of the selected one of the blocks 50, using a Select/Boost instruction, selects a word-line at a first timing, and transfers data from a memory cell connected to the selected word-line to the sense amplifier 58 at a second timing four cycles after the first timing. At a third timing, the data is read out by a Read-1 instruction. Since the burst length is 2 in this case, data is read out two times consecutively, so that a stream of 8 bits of data read out during 8 cycles of the clock are output as data outputs 1 through 17. The local row scheduler 51 executes a Precharge instruction after the Read instruction. By doing so, bit-lines are precharged, and the selected one of the blocks 50 is back in the ready condition.

The global row scheduler 41 selects one of the blocks 50 one after another at every other timing (8-clock intervals) as shown in FIGS. 4A through 4L. At a given timing, a first block (e.g., block 1) of the blocks 50 executes the Select/Boost instruction. Two timings later, the first block executes the Read instruction, and, at the same time, a second block (block 2) of the blocks 50 is selected to carry out the Select/Boost instruction. After further two timings, the first block executes the Precharge instruction, the second block executes the Read instruction, and a third block (block 3) of the blocks 50 is selected to execute the Select/Boost instruction. In this manner, a different one of the blocks 50 is successively selected to carry out the Select/Boost instruction, the Read instruction, and the Precharge instruction. As a result, the blocks 50 operate as in the pipeline operation to allow consecutive reading of data.

With reference to FIGS. 4A through 4L, a set of the Read-1 and Read-2 instructions is carried out by only one of the blocks 50 at a given time. This is because the bus 60 and the data-transfer/column decoder 46 are shared by the plurality of the blocks 50 in FIG. 3. Namely, if two or more blocks simultaneously attempt the read operation, competing demands are created for the bus 60 and the data-transfer/column decoder 46 so that a normal operation cannot be guaranteed. In FIGS. 4A through 4L, the blocks 1 through 5 are in operation simultaneously, but only one of these blocks is allowed to carry out the Read instruction at a given time. On the other hand, an operation (e.g., standby operation) for which no competing demand is created by a plurality of blocks is created can be simultaneously executed by more than one block. Depending on data addresses to be accessed, the same block may be selected consecutively. In this case, a continuous pipeline operation is disturbed.

Although the schedule of each block shown in FIGS. 4A through 4L has been described with reference to the data-read operation, it is apparent that the data-write operation for the DRAM can be carried out in the same schedule. In this case, a Write instruction replaces the Read instruction to write data in one of the memory arrays 56 and 57 via the sense amplifier 58. In the data-write operation, the Select/Boost operation only selects a word-line without a need to transfer data into the sense amplifier 58.

In each of the blocks 50, a word-line corresponding to the row address supplied at the time of the block selection is chosen. If a row address supplied to a selected one of the blocks 50 was changed to another row address supplied to another one of the blocks 50 before the completion of the data-read operation in the former one of the blocks 50, it would cause a problem. Thus, each of the blocks 50 should be provided with the row-address register 52 serving as a latch for storing a row address supplied thereto. This latch may be provided for each word-line to store a decoded address. Alternately, this latch may be provided as a predecode register for storing an address prior to decoding. Alternately, this latch may be provided as a register to store an address of an intermediate representation.

The DRAM chip according to the present invention receives a 16-bit command/address input at instruction, etc., at the predetermined timing according to the supplied command indicating the 8-burst operation, the 16-burst operation, or the like.

FIGS. 6A through 6H are time charts showing relations between the Flag-0 input, the command/address input, and other signals. The Flag-0 signal and the command/address input are supplied to the chip in synchronism with the strobe signal input to the chip, as shown in FIGS. 6A through 6D. The command/address input is comprised of 4 cycles of the input strobe signal (see FIG. 5A), and is read by the chip using a rising edge of the flag signal Flag-0 as read-start timing. Here, in the present invention, 1/2 cycle of the input strobe signal is used in actuality as one cycle of this reading operation for convenience. In FIGS. 6A through 6D, the command address input is provided with a delay by one cycle (1/2 cycle of the strobe signal) compared to the flag input. This is because the flag signal needs to be input ahead of the command/address input.

FIGS. 6E through 6H shows examples of internal clocks Φ1 through Φ4, which are used for reading the command/address input of FIG. 6D. The internal clocks Φ1 through 44 as shown in the figures are generated by the internal-clock generator 23 of FIG. 3, and are supplied to input-signal buffer 11 of FIG. 3. With respect to each signal line of input signals, the input-signal buffer 11 of FIG. 3 may include four input buffers which receive the same signal on the same signal line, and latches this signal by using the internal clocks Φ1 through Φ4 at four consecutive cycles. In this manner, these four input buffers can demultiplex an input packet shown in FIG. 5A. As for detailed circuit structures of the input-signal buffer 11 and the internal-clock generator 23 are well within the scope of ordinary skill in the art, and, thus, a description thereof will be omitted. the node IN of FIG. 3. This command/address input is provided at the time when a flag signal Flag-0 input to the CTRL node of the chip is at a high level. The command/address input is demultiplexed in synchronism with the strobe signal during the high level of the flag signal Flag-0 to be broken into four different contents.

FIGS. 5A and 5B are table charts showing the four contents of the command/address input. As shown in FIG. 5A, the command/address input contains a device ID, a bank address, and a command during a first cycle of the strobe signal (Strobe). During a second cycle of the strobe signal, the command/address input includes the row address and the block address. Further, the column address is provided in the command/address input during a third cycle of the strobe signal. The command/address input during the last and fourth cycle of the strobe signal is not currently used.

As shown in FIG. 5A, the command/address input is supplied to the memory chip in a packet format by multiplexing the addresses and the command.

FIG. 5B shows details of the command which is defined by bits 8 through 15 of the command/address input during the first cycle. As shown in FIG. 5B, for example, the bit 1 being "1" indicates the refresh operation, the bit 2 being "1" indicates the selection of a word-line, and the bits 4 through 6 specify a burst length. In this manner, the command specified by a bit pattern defines an operation of one of the blocks 50. Namely, the command specified by the bit pattern is supplied to one of the blocks 50 indicated by the block address, in a bank indicated by the bank address, in the device indicated by the device ID, as shown in the command/address input of FIG. 5A. In response, one of the blocks 50 carries out the Select/Boost instruction, the Read instruction, the Precharge

The internal clock Φ4 is also used as a signal for indicating a start timing of operations to internal circuits in the DRAM chip. Namely, when this signal is activated, the internal circuits such as the global row scheduler 41, etc., start operations.

Having described the command/address input, the demultiplexing of the command/address input, and the generation of the internal clocks, configurations and operations of the global row scheduler 41, the column scheduler 44, and the local row scheduler 51 of FIG. 3 will be described below in detail.

First, an internal clock θ1, which is used for synchronizing the operations of the global row scheduler 41, the column scheduler 44, and the local row scheduler 51 will be described. FIGS. 7A through 7D are time charts showing the strobe signal, the internal clock Φ4, an internal clock Θ1, and an inverted internal clock /Θ1, respectively. As previously described, the operations of the global row scheduler 41, etc., are started by using the signal Φ4 of FIG. 6H as a start signal. As shown in FIGS. 7A through 7D, the internal clock Θ1 switches between HIGH and LOW every two cycles one cycle after the signal Φ4 changes to LOW. A circuit configuration for generating the internal clocks Θ1 and /Θ1 is well within the range of ordinary skill in the art, and a description thereof will be omitted.

In this first embodiment of the present invention, each of the blocks 50 changes its operation state every four cycles, so that the signal Θ1 has a period matching with the timing of the operation-state transition. That is, this signal Θ1 can be used as the internal clock for operating the global row scheduler 41, the column scheduler 44, and the local row scheduler 51. The signal Θ1 can be generated based on the strobe signal-Strobe and the signal Φ4 (or Φ3 or the like) by using knowledge well within the ordinary skill in the field, and a description of a circuit for generating the signal Θ1 will be omitted.

FIG. 8 and FIG. 9 are circuit diagrams showing a configuration of the global row scheduler 41. FIG. 8 shows a part which selects one of the blocks 50 and specifies a burst length, etc., for the selected one of the blocks 50. FIG. 9 shows a part which selects a row address for the selected one of the blocks 50.

As shown in FIG. 8, the global row scheduler 41 includes a plurality of block-selection MS (master-slave) flip-flops 201, a plurality of command MS flip-flops 202, auto-precharge MS flip-flops 203 and 204, a decoding unit 205, a driving unit 206, and a signal-output-timing MS flip-flop 207.

The mode-selection circuits 200 receive the decoded refresh-mode signal, the decoded normal-mode signal, etc., from the mode decoder 17 of FIG. 3, and, also, receive address signals X9, /X9, . . . , X11, and /X11 for selecting one of the blocks 50 from the address pre-decoder 13. Also, the mode-selection circuits 200 receive address signals X9C, X10C, and X11C for selecting one of the blocks 50 at the time of the refresh operation from the refresh-address counter 42. Each of the mode-selection circuits 200 includes NMOS transistors 211 through 214 and an inverter 215. One of the mode-selection circuits 200 for receiving the address signal X9, for example, supplies an inverted address signal /X9C to one of the block-selection MS flip-flops 201 when the refresh-mode signal is ON, and supplies address signals X9 and /X9 to the corresponding block-selection MS flip-flops 201 when the normal-mode signal is ON.

The block-selection MS flip-flops 201 latch the supplied data by using the internal clock Θ1 and the inverted internal clock /Θ1. Output signals from the block-selection MS flip-flops 201 for selecting one of the blocks 50 are supplied to the decoding unit 205.

The decoding unit 205 includes a plurality of NAND circuits 221, a plurality of NAND circuits 223, a plurality of inverters 222, and a plurality of inverters 224. The decoding unit 205 receives the Select signal from the signal-output-timing MS flip-flop 207, and outputs a selection signal for the blocks 50 when the Select signal is HIGH. When (X9, X10, X11) is (1, 1, 0), for example, only a selection signal supplied to a third one of the blocks 50 becomes HIGH.

Each of the command MS flip-flops 202 receives a signal indicating a burst type, a signal indicating a compulsory precharge, or the like from the command decoder 15, and latches the received signal. The latched signal is supplied to each of the blocks 50 via the driving unit 206 comprised of inverters 231.

The auto-precharge MS flip-flop 203 receives a signal indicating an auto-precharge from the command decoder 15, and latches the received signal. The latched signal is supplied to each of the blocks 50 via the driving unit 206 after a further latch by the auto-precharge MS flip-flop 204 for the timing adjustment.

In this manner, the global row scheduler 41 supplies the selection signal and the command for indicating an operation to the selected one of the blocks 50.

As shown in FIG. 9, the global row scheduler 41 further includes a plurality of mode-selection circuits 200, a plurality of row-address-selection MS flip-flops 240, a decoding unit 241, and a row-address-selection gate unit 242.

The row-address-mode-selection circuits 200 are the same circuits as the mode-selection circuits 200 of FIG. 8, and receive a signal for indicating a mode, address signals X0 through X8 and /X0 through /X8 for indicating a row address, and address signals X0C through X8C for indicating a refresh-mode row address. Then, the row-address-mode-selection circuits 200 select appropriate address signals matching with the indicated mode. The row-address-selection MS flip-flops 240 latch the signals from the row-address-mode-selection circuits 200 by using the internal clock Θ1 and the inverse internal clock /Θ1. The decoding unit 241 is a circuit of the same type as the decoding unit 205 of FIG. 8, and decodes address signals from the row-address-selection MS flip-flops 240 to output a decoded address signal. This decoded address signal is output for one cycle after the input of the Select signal. The row-address-selection gate unit 242 outputs the most significant bit X8 of the row address and the most significant bit /X8 of the inverted row address for one cycle after the input of the Select signal.

In this manner, the global row scheduler 41 supplies to the selected one of the blocks 50 the row address in addition to the selection signal and the command for indicating the operation.

FIG. 10 is a circuit diagram of the column scheduler 44. The column scheduler 44 of FIG. 10 includes a plurality of column-address-mode-selection circuits 200, a plurality of column-address-selection MS flip-flops 240, a decoding unit 241, a column-address-selection gate unit 242, and the signal-output-timing MS flip-flop 207. The configuration of the column scheduler 44 of FIG. 10 is the same as that of the row-address-selection portion of the global row scheduler 41 shown in FIG. 9. The only difference is that the input address signals are not row address signals as in FIG. 9 but column address signals Y0 through Y6 and /Y0 through /Y6 from the address pre-decoder 13 and column address signals Y0C through Y6C from the burst-address counter 45. Since the circuit configuration and the operation of FIG. 10 are the same as those of FIG. 9, a description thereof will be omitted.

FIG. 11 is a circuit diagram of the local row scheduler 51. As shown in FIG. 11, the local row scheduler 51 includes an illegal-input checking unit 401, a word-line-selection unit 402, a read-signal-generation unit 403, a precharge-signal-generation unit 404, a precharge-mode-operation unit 405, and a plurality of MS flip-flops 406 through 409.

The MS flip-flops 406 through 409 receive signals indicating the 16-burst length, the 8-burst length, the 4-burst length, and the compulsory-burst operation, and latch these signals. Outputs of the MS flip-flops 406 through 409 are used by the read-signal-generation unit 403, the precharge-signal-generation unit 404, and the precharge-mode-operation unit 405.

The illegal-input checking unit 401 includes NAND circuits 421 and 422 and inverters 423 and 424. The illegal-input checking unit 401 detects a condition such as consecutive accesses to the same one of the blocks 50 which causes a disturbance of the pipeline operation, and outputs an illegal-input signal indicating a detection of an illegal input. In FIG. 11, a circuit for generating an illegal-input signal when the same one of the blocks 50 is selected during a precharge operation is shown as an example.

The illegal-input checking unit 401 receives the block selection signal for selecting the pertinent one of the blocks 50 from the global row scheduler 41. A precharge signal internally generated to be ON during the precharge operation and the block selection signal are input to the NAND circuit 421. An output of the NAND circuit 421 is LOW only when the block selection is made during the precharge operation, so as to indicate the detection of an illegal input. Based on this output signal, a proper measure can be taken such as making the global row scheduler 41 delay the block selection or making the global row scheduler 41 ignore the block selection of the pertinent block.

In the illegal-input checking unit 401, the NAND circuit 422 receives an inversion of the precharge signal from the inverter 423 and the block selection signal. Thus, the inverter 424 connected to an output of the NAND circuit 422 supplies the block selection signal to the word-line-selection unit 402 only when the precharge operation is not in process.

The word-line-selection unit 402, receiving the block selection signal which is HIGH after the first cycle of the internal clock Θ1, outputs a word-line-selection signal during the first two cycle of the internal clock Θ1 (corresponding to 8 cycles of the strobe signal Strobe) to select a word-line (row address). The word-line-selection unit 402 includes MS flip-flops 431 and 432, a NOR circuit 433, and an inverter 434. The MS flip-flops 431 and 432 form a shift register which shifts the block selection signal to the right at each cycle based on the internal clock Θ1 (and /Θ1). Thus, the NOR circuit 433 receiving outputs of the MS flip-flops 431 and 432 generates a LOW output during the first two cycles. Thus, the inverter 434 outputs the word-line-selection signal which is HIGH during the first two cycles.

The reason why the word-line-selection signal is HIGH for the first two cycles is because the word-line-selection operation (Select operation) and the operation for transferring data to the sense amplifier 58 (Boost operation) require two cycles in total, as shown in the schedules of FIGS. 4A through 4L.

The read-signal-generation unit 403 receives a HIGH signal during the second cycle from the word-line-selection unit 402, and outputs a read signal starting from a third cycle for a period variable according to the burst type. The read-signal-generation unit 403 includes MS flip-flops 441 through 444, NAND circuits 445 through 450, a NOR circuit 451, inverters 452 through 454, and a NOR circuit 455.

The MS flip-flop 441 receives a HIGH signal during the second cycle,-and generates a HIGH signal during the third cycle. The HIGH signal during the third cycle is provided to the NAND circuits 446, 449, and 450.

In the period before the third cycle, outputs of the NAND circuits 446, 449, and 450 are HIGH since the output of the MS flip-flop 441 is LOW. Thus, an output of the inverter 452 is-LOW. Since an output of the MS flip-flop 442 is LOW, an output of the NOR circuit 451 is HIGH. Also, outputs of the MS flip-flops 443 and 444 are LOW, so that outputs of the inverters 453 and 454 are HIGH, thereby resulting in both outputs of the NAND circuits 447 and 448 being LOW. Thus, an output of the NOR circuit 455 is HIGH. As a result, in the period prior to the third cycle, two inputs of the NAND circuit 445 are HIGH, so that the output of the read-signal-generation unit 403 is LOW. Only when at least one of these two inputs becomes LOW, will the output of the read-signal-generation unit 403 become HIGH.

In the third cycle, the output of the NAND circuit 446 becomes LOW when the 16-burst length is selected. Alternately, the output of the NAND circuit 449 becomes LOW when the 8-burst length is selected. Alternately, the output of the NAND circuit 450 becomes LOW when the 4-burst length is selected. Therefore, irrespective of the selected burst mode, the read signal becomes HIGH during the third cycle.

When the 4-burst mode is selected, the output of the NAND circuit 448 becomes HIGH at the third cycle. This output is read by the MS flip-flop 461 at an end of the operation of the read-signal-generation unit 403. Namely, the read-signal-generation unit 403 generates the read signal which is HIGH only during the third cycle.

When the 8-burst mode is selected, the output of the NAND circuit 447 becomes HIGH during the third cycle, so that the output of the MS flip-flop 444 becomes HIGH during a fourth cycle. The output of the MS flip-flop 444 makes the output of the NAND circuit 448 HIGH via the inverter 454. Therefore, the output of the NOR circuit 455 becomes LOW, so that the read-signal-generation unit 403 generates a HIGH signal during the fourth cycle. The HIGH output of the NAND circuit 448 is read by the MS flip-flop 461 at an end of the operation of the read-signal-generation unit 403. Namely, the read-signal-generation unit 403 generates the read signal which is HIGH during the third cycle and the fourth cycle.

When 16-burst mode is selected, the output of the inverter 452 becomes HIGH during the third cycle, so that the output of the MS flip-flop 442 becomes HIGH during the fourth cycle. Therefore, the output of the NOR circuit 451 becomes LOW, so that the read-signal-generation unit 403 outputs a HIGH signal during the fourth cycle. By the HIGH output of the MS flip-flop 442, the output of the MS flip-flop 443 becomes HIGH during a fifth cycle. The HIGH output of the MS flip-flop 443 makes the output of the NAND circuit 447 HIGH. As a result, the output of the NOR circuit 455 becomes LOW, and the read-signal-generation unit 403 outputs a HIGH signal during the fifth cycle. The HIGH output of the NAND circuit 447 is input to the MS flip-flop 444, so that the output of the MS flip-flop 444 becomes HIGH during a sixth cycle. The output of the MS flip-flop 444 makes the output of the NAND circuit 448 HIGH via the inverter 454. With the HIGH output of the NAND circuit 448, the output of the NOR circuit 455 becomes LOW. Thus, the read-signal-generation unit 403 outputs a HIGH signal during the sixth cycle. The HIGH output of the NAND circuit 448 is read by the MS flip-flop 461 at an end of the operation of the read-signal-generation unit 403. Namely, the read-signal-generation unit 403 generates the read signal which is HIGH during the third cycle, the fourth cycle, the fifth cycle, and the sixth cycle.

The precharge-signal-generation unit 404 generates a precharge signal at a predetermined timing when the precharge mode is selected. The HIGH signal supplied to the MS flip-flop 461 at a given cycle is latched by the MS flip-flop 461 one cycle later to generate a HIGH output. The NAND circuit 464 receiving the output of the MS flip-flop 461 at one input thereof receives an output of the precharge-mode-operation unit 405 at the other input thereof. When both inputs are HIGH, the output of the NAND circuit 464 becomes LOW. In this case, the output of the NAND circuit 463 becomes HIGH, so that the output of the precharge-signal-generation unit 404 becomes HIGH. The HIGH output of the NAND circuit 463 is read by the MS flip-flop 462, and, thus, the precharge-signal-generation unit 404 generates a HIGH output also during a next cycle.

Namely, the precharge operation is carried out during the fourth and fifth cycles when the 4-burst mode is selected. Alternately, the precharge operation is carried out during the fifth and sixth cycles when the 8-burst mode is selected. Alternately, the precharge operation is carried out during the seventh and eighth cycles when the 16-burst mode is selected.

The precharge-mode-operation unit 405 is used for initiating the precharge operation when the precharge mode is selected. The precharge-mode-operation unit 405 includes MS flip-flops 471 through 477, NAND circuits 478 through 482, and inverters 483 through 485. The MS flip-flops 471 through 477 constitutes a shift register which successively shifts a supplied precharge signal to the right. Thus, the output of the MS flip-flop 473 becomes HIGH during the third cycle.

When the 4-burst mode is indicated, the output of the NAND circuit 480 becomes LOW during the third cycle, so that the output of the NAND circuit 482 becomes HIGH. Thus, the output of the MS flip-flop 477 becomes HIGH during the fourth cycle. The output of the MS flip-flop 477 is supplied to one input of the NAND circuit 464 of the precharge-signal-generation unit 404. When the output of the MS flip-flop 477 becomes HIGH during the fourth cycle, the other input of the NAND circuit 464 also becomes HIGH, so that the output of the NAND circuit 464 becomes LOW. As a result, the output of the precharge-signal-generation unit 404 becomes HIGH during the fourth cycle as previously described.

When the 8-burst operation is indicated, the output of the NAND circuit 479 becomes LOW during the third cycle, so that the output of the NAND circuit 481 becomes HIGH. Thus, the output of the MS flip-flop 476 becomes HIGH during the fourth cycle. The output of the MS flip-flop 476 is supplied to the MS flip-flop 477 via the inverter 485 and the NAND circuit 482. As a result, the output of the MS flip-flop 477 becomes HIGH during the fifth cycle. During the fifth cycle, therefore, the output of the precharge-signal-generation unit 404 becomes HIGH.

When the 16-burst operation is selected, the output of the NAND circuit 478 becomes LOW during the third cycle, so that the output of the inverter 483 becomes HIGH. Thus, the output of the MS flip-flop 474 becomes HIGH during the fourth cycle. Then, the output of the MS flip-flop 475 becomes HIGH during the fifth cycle. The output of the MS flip-flop 475 is supplied to the MS flip-flop 476 via the inverter 484 and the NAND circuit 481. As a result, the output of the MS flip-flop 476 becomes HIGH during the sixth cycle. The output of the MS flip-flop 476 is supplied to the MS flip-flop 477 via the inverter 485 and the NAND circuit 482. The output of the MS flip-flop 477 becomes HIGH during the seventh cycle. Accordingly, the output of the precharge-signal-generation unit 404 becomes HIGH during the seventh cycle.

In this manner, when one of the blocks 50 is selected, the local row scheduler 51 of the selected block carries out the word-line-selection operation, the data-read operation, and the precharge operation. Also, the local row scheduler 51 controls the operation timings of the above operations based on the selected burst type, the selected precharge type, etc.

The local row scheduler 51 of FIG. 11 is merely an example, and other variations may be devised. Also, it is apparent that the data-write operation can be carried out in the same timing as that of the data-read operation described above. Further, the functions of the local row scheduler 51 may be included in the global row scheduler 41. In this case, a DRAM using the configuration of FIG. 2B according to the present invention is provided. This variation is achieved by incorporating the circuit of the local row scheduler 51 into the global row scheduler 41 and by providing control lines for conveying the operation instructions to each of the blocks 50. Modifications necessary to achieve this variation are well within the scope of ordinary skill in the art, and a description thereof will be omitted.

FIG. 12 is a circuit diagram of a circuit for the parallel-to-serial conversion for the chip output. FIG. 13A through 13E are time charts of signals used in the circuit of FIG. 12. In the first embodiment of DRAM according to the present invention, a data-read operation reads 4 bits of data in parallel, and converts it into serial data at the time of the data output, as previously described with reference to the schedule of each block shown in FIGS. 4A through 4L. The circuit of FIG. 12 is an interface circuit for converting the 4-bit data from parallel data to serial data so as to output the data from the DRAM chip of the present invention.

The circuit of FIG. 12 includes tri-state buffers 520 and an output buffer 521. Each of the tri-state buffers 520 receives data via the 4-bit-parallel data bus, and is provided with one of internal clocks H, I, J, and K, shown in FIGS. 13B through 13E, triggered by the signal Φ3 of FIG. 13A. A circuit well within the scope of ordinary skill in the art can generate the internal clocks H, I, J, and K by using the signal 43 as a trigger, and a description of such a circuit structure will be omitted.

In each of the tri-state buffers 520, when the supplied internal clock is HIGH, an output of a NAND circuit 524 becomes HIGH, and an output of a NOR circuit 525 becomes LOW, so that the transistors 522 and 523 are turned off. In this condition, the output of the tri-state buffer 520 is floating. When the supplied internal clock is LOW, the tri-state buffer 520 passes the input data from the data bus to the output thereof. The output buffer 521, using a strobe signal having the same phase as the input strobe signal, holds the data supplied from the tri-state buffers 520. Accordingly, each time the internal clocks H, I, J, and K successively become LOW, the tri-state buffers 520 sequentially supply the data to the output buffer 521, which outputs the data from the chip. In this manner, internal parallel data is multiplexed into serial data in the output buffer 521 to form an output packet of the DRAM chip according the present invention.

A plurality of the circuits of FIG. 12 are provided to correspond to the number of the output pins of the chip. For example, if the number of the output pins of the DRAM of the present invention is 16, 16 circuits the same as the circuit of FIG. 12 are provided. By the same token, if the number of input pins of the DRAM is 16, 16 circuits the same as the circuit of FIG. 12 are provided.

The above-described embodiment of the DRAM according to the present invention may be applied to a conventional SDRAM. FIGS. 14A through 14L are illustrative drawings showing schedules of blocks of the SDRAM according to the present invention when the burst length is 1 and the CAS latency is 1.

Each block is initially in the ready condition. When the global row scheduler 41 selects one of the blocks 50, the local row scheduler 51 in the selected one of the blocks 50 selects a word-line at a first cycle of the internal clock Φ1 by using the Select/Boost instruction. At a second cycle, data stored in a memory cell connected to the selected word-line is transferred to the sense amplifier 58. At a third cycle, the data is read by the Read instruction. Since the burst length is 1 in this case, the reading of the data is carried out only once, with 4 bits of the data being successively read out during 4 cycles of the strobe signal. The local row scheduler 51 carries out the Precharge instruction after the Read instruction. The bit-lines are thus precharged to bring the selected one of the blocks 50 to the ready condition.

Further, the present invention is not limited to these embodiments, but various variations and modifications may be made without departing from the scope of the present invention.

Patent Citations
Cited PatentFiling datePublication dateApplicantTitle
US4633434 *Apr 2, 1984Dec 30, 1986Sperry CorporationHigh performance storage unit
US4680730 *Jul 9, 1984Jul 14, 1987Hitachi, Ltd.Storage control apparatus
US5060145 *Sep 6, 1989Oct 22, 1991Unisys CorporationMemory access system for pipelined data paths to and from storage
US5384745 *Apr 14, 1993Jan 24, 1995Mitsubishi Denki Kabushiki KaishaSynchronous semiconductor memory device
US5410670 *Jun 2, 1993Apr 25, 1995Microunity Systems Engineering, Inc.Accessing system that reduces access times due to transmission delays and I/O access circuitry in a burst mode random access memory
US5737563 *Jun 7, 1995Apr 7, 1998Texas Instruments IncorporatedDetermination of memory bank sizes in a computer system
US5745118 *Jun 6, 1995Apr 28, 1998Hewlett-Packard Company3D bypass for download of textures
US5748914 *Oct 19, 1995May 5, 1998Rambus, Inc.Protocol for communication with dynamic memory
US5768559 *Apr 22, 1996Jun 16, 1998Fujitsu LimitedMultiple bank structured memory access device having flexible setting of a pipeline stage number
EP0384620A2 *Feb 9, 1990Aug 29, 1990Data General CorporationHigh performance memory system
EP0498525A2 *Jan 6, 1992Aug 12, 1992Advanced Micro Devices, Inc.Sequential memory accessing
WO1988009995A1 *Apr 22, 1988Dec 15, 1988Hughes Aircraft CoPipeline memory structure
Referenced by
Citing PatentFiling datePublication dateApplicantTitle
US6175905Jul 30, 1998Jan 16, 2001Micron Technology, Inc.Method and system for bypassing pipelines in a pipelined memory command generator
US6178488Aug 27, 1998Jan 23, 2001Micron Technology, Inc.Method and apparatus for processing pipelined memory commands
US6230245Feb 11, 1997May 8, 2001Micron Technology, Inc.Method and apparatus for generating a variable sequence of memory device command signals
US6286062Jul 1, 1997Sep 4, 2001Micron Technology, Inc.Pipelined packet-oriented memory system having a unidirectional command and address bus and a bidirectional data bus
US6292877 *Nov 5, 1999Sep 18, 2001Micron Technology, Inc.Plural pipelined packet-oriented memory systems having a unidirectional command and address bus and a bidirectional data bus
US6347354 *Oct 9, 1998Feb 12, 2002Rambus IncorporatedApparatus and method for maximizing information transfers over limited interconnect resources
US6360292Oct 10, 2000Mar 19, 2002Micron Technology, Inc.Method and system for processing pipelined memory commands
US6366992Dec 26, 2000Apr 2, 2002Micron Technology, Inc.Method and system for bypassing pipelines in a pipelined memory command generator
US6370627Jul 25, 2000Apr 9, 2002Micron Technology, Inc.Memory device command buffer apparatus and method and memory devices and computer systems using same
US6418495Nov 5, 1999Jul 9, 2002Micron Technology, Inc.Pipelined packet-oriented memory system having a unidirectional command and address bus and a bidirectional data bus
US6484244Jun 17, 1997Nov 19, 2002Micron Technology, Inc.Method and system for storing and processing multiple memory commands
US6519675Nov 22, 1999Feb 11, 2003Micron Technology, Inc.Two step memory device command buffer apparatus and method and memory devices and computer systems using same
US6542569Mar 8, 2001Apr 1, 2003Micron Technology, Inc.Memory device command buffer apparatus and method and memory devices and computer systems using same
US6622203 *May 29, 2001Sep 16, 2003Agilent Technologies, Inc.Embedded memory access method and system for application specific integrated circuits
US6636443 *Oct 30, 2001Oct 21, 2003Hynix Semiconductor Inc.Semiconductor memory device having row buffers
US6708248Dec 8, 1999Mar 16, 2004Rambus Inc.Memory system with channel multiplexing of multiple memory devices
US6708262Mar 19, 2001Mar 16, 2004Micron Technology, Inc.Memory device command signal generator
US6721227 *Feb 11, 2002Apr 13, 2004Micron Technology, Inc.User selectable banks for DRAM
US6748442 *Jan 3, 2000Jun 8, 2004Advanced Micro Devices, Inc.Method and apparatus for using a control signal on a packet based communication link
US6750910 *Jul 15, 1999Jun 15, 2004Texas Instruments IncorporatedOptical black and offset correction in CCD signal processing
US6757789Jan 30, 2002Jun 29, 2004Rambus, Inc.Apparatus and method for maximizing information transfers over limited interconnect resources
US6804743Nov 6, 2002Oct 12, 2004Micron Technology, Inc.Two step memory device command buffer apparatus and method and memory devices and computer systems using same
US6829184 *Jan 28, 2002Dec 7, 2004Intel CorporationApparatus and method for encoding auto-precharge
US6900812 *Aug 2, 2000May 31, 2005Ati International SrlLogic enhanced memory and method therefore
US6963949Nov 5, 1999Nov 8, 2005Micron Technology, Inc.Pipelined packet-oriented memory system having a unidirectional command and address bus and a bidirectional data bus
US7012613May 2, 2000Mar 14, 2006Ati International SrlMethod and apparatus for fragment scriptor for use in over-sampling anti-aliasing
US7020740 *Dec 6, 2000Mar 28, 2006Sun Microsystems, Inc.Computer arrangement using non-refreshed DRAM
US7035962Nov 5, 1999Apr 25, 2006Micron Technology, Inc.Pipelined packet-oriented memory system having a unidirectional command and address bus and a bidirectional data bus
US7039782Oct 20, 2003May 2, 2006Rambus Inc.Memory system with channel multiplexing of multiple memory devices
US7080261Dec 7, 1999Jul 18, 2006Sun Microsystems, Inc.Computer-readable medium with microprocessor to control reading and computer arranged to communicate with such a medium
US7145991Jan 9, 2004Dec 5, 2006Sun Microsystem, Inc.Superposition of data over voice
US7185110Apr 16, 2002Feb 27, 2007Sun Microsystems, Inc.Data exchange system comprising portable data processing units
US7197596Feb 2, 2006Mar 27, 2007Sun Microsystems, Inc.Computer arrangement using non-refreshed DRAM
US7203122Feb 18, 2004Apr 10, 2007Micron Technology, Inc.User selectable banks for DRAM
US7273169Oct 19, 2004Sep 25, 2007Sun Microsystems, Inc.Secure photo carrying identification device, as well as means and method for authenticating such an identification device
US7360039Jun 21, 2004Apr 15, 2008Belle Gate Investment B.V.Arrangements storing different versions of a set of data in separate memory areas and method for updating a set of data in a memory
US7409558 *Sep 2, 2004Aug 5, 2008International Business Machines CorporationLow-latency data decryption interface
US7636271Apr 10, 2007Dec 22, 2009Micron Technology, Inc.User selectable banks for DRAM
US7719916 *Dec 31, 2007May 18, 2010Hynix Semiconductor, Inc.Semiconductor memory device
US7724594Mar 7, 2008May 25, 2010Hynix Semiconductor Inc.Leakage current control device of semiconductor memory device
US7828218Jul 20, 2000Nov 9, 2010Oracle America, Inc.Method and system of communicating devices, and devices therefor, with protected data transfer
US7995420Dec 22, 2009Aug 9, 2011Round Rock Research, LlcUser selectable banks for DRAM
US8006032Aug 22, 2007Aug 23, 2011Globalfoundries Inc.Optimal solution to control data channels
US8069353 *Jun 19, 2008Nov 29, 2011International Business Machines CorporationLow-latency data decryption interface
WO2001042929A1 *Nov 24, 2000Jun 14, 2001Rambus IncMemory system with channel multiplexing of multiple memory devices
Classifications
U.S. Classification711/169, 711/5, 711/168, 711/167
International ClassificationG11C29/00, G11C11/407, G11C11/401, G11C7/00, G11C7/10
Cooperative ClassificationG11C7/1039, G11C7/1072, G11C29/88, G11C7/1015
European ClassificationG11C29/88, G11C7/10S, G11C7/10M5, G11C7/10M
Legal Events
DateCodeEventDescription
Sep 14, 2011FPAYFee payment
Year of fee payment: 12
Jul 22, 2010ASAssignment
Free format text: CHANGE OF NAME;ASSIGNOR:FUJITSU MICROELECTRONICS LIMITED;REEL/FRAME:024982/0245
Effective date: 20100401
Owner name: FUJITSU SEMICONDUCTOR LIMITED, JAPAN
Dec 12, 2008ASAssignment
Owner name: FUJITSU MICROELECTRONICS LIMITED, JAPAN
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:FUJITSU LIMITED;REEL/FRAME:021998/0645
Effective date: 20081104
Owner name: FUJITSU MICROELECTRONICS LIMITED,JAPAN
Sep 17, 2007FPAYFee payment
Year of fee payment: 8
Sep 29, 2003FPAYFee payment
Year of fee payment: 4
May 30, 1997ASAssignment
Owner name: FUJITSU LIMITED, JAPAN
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:OKAJIMA, YOSHINORI;REEL/FRAME:008555/0483
Effective date: 19970509